Professional Documents
Culture Documents
E
H ERB BLANK nvironmental, social, and corpo- not a random process as operationalized by
is the head of business rate governance (ESG) investing Adler and Kritzman [2008]—who used
development at S-Network
has exploded. As of January 2015, Monte Carlo simulations to argue that SRI
Global Indexes, Inc., in
New York, NY. investing in ESG portfolios had incurs opportunity cost—screening that
hblank@snetworkinc.com increased 76% since 2012 and accounted excludes and includes companies based solely
for more than one-sixth of all investing on their industries and products cannot be
GREGG SGAMBATI (McQuillen [2015]). As the more conscien- expected to work as a reliably profitable
is the head of ESG Solu- tious approaches to investing have evolved strategy. According to a thorough analysis by
tions at S-Network Global
Indexes, Inc., in New
from simple avoidance of “sin stocks” (such as Renneboog, Ter Horst, and Zhang [2008a],
York, NY. those in tobacco and alcohol) to increasingly as of December 2003, investors were in tran-
gsgambati@snetworkinc.com nuanced and performance-driven strategies, sition from simple negative screening to
demand has grown for reliable data. While more ESG-oriented methods. Globally, the
ZACK T RUELSON popular interest in such strategies has been screen ratios were mixed, but particularly
is a marketing associate
on the rise for decades, their focus and ter- in the United States, negative screens (97%)
at S-Network Global
Indexes, Inc., in New minology have shifted away from “ethical dominated positive (69%) and anti-“sin”
York, NY. investing” and toward “socially responsible screens (92%) outweighed environmental
ztruelson@snetworkinc.com investing” (SRI) and similar names (Eccles (72%) and social/governance (68%), with
and Viviers [2011]). From the beginning, the ethical screens comparable at 57%. (These
trend has been somewhere under the collec- are the percentages of funds that employed
tive umbrella of ESG issues, but recently, we each type, so they sum to more than 100%.)
have seen attention more broadly distributed And not surprisingly, results were likewise
to cover the full scope of ESG. Instead of mediocre: another analysis (by the same
using such a blunt instrument as negative authors) of socially responsible (SR) funds
screening at the industry level, more investors from 1992 to 2003 found them to under-
are paying attention to what distinguishes perform (Renneboog, Ter Horst, and Zhang
companies within industries, opening an array [2008b]).
of new inquiries about how—rather than just More recent findings are still mixed.
what—the companies produce. Humphrey and Tan [2013] found no effect
ESG/SRI-oriented methods have been on returns from either positive or negative
around for several decades, dating back at screens, while Trinks and Scholtens [2015]
least to the launch of the Calvert Social found a significant opportunity cost for nega-
Investment Fund in 1982. It was not always tive screening. Statman and Glushkov [2008]
as data-driven as it is today. Although it is found that negative screening of “shunned”
model the behavior of the portfolio manager. The study conceived rating systems.1 These ratings provide a quan-
by Adler and Kritzman modeled the stock selection as titative grounding for the purposes of both research and
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
a randomized negative screen, essentially excluding investing, whether aiming at the whole of ESG or at the
a given percentage of the universe of stocks without individual ESG components (which are referred to as
regard to any of the stocks’ attributes. De and Clayman “pillars” in the TRCRR system).
[2015] also used a method of randomized portfolio cre- As detailed by Blank [2013], the development of
ation, but did so after restricting the universe of stocks the TRCRR involved hundreds of primary researchers
according to Thomson Reuters/ASSET4 ESG ratings. and a lot of careful thought, resulting in different designs
They found that in real-world practice, there is clearly for the three ESG pillars. The goal was to take a vast
room for improvement. Wimmer [2013] found that array of information and distil it into numerical rat-
the average ESG ratings of the portfolios of SR mutual ings that could be used to fairly compare companies
funds persist for only two to three years—but rather across and within industries and geographical regions.
than being caused by high volatility in the ESG ratings Achieving that goal required numerous steps of con-
themselves, the short time horizon is caused by turnover verting, weighting, and normalizing scores, as well as
in the portfolios. In other words, it is not that highly qualitative decisions about how to group companies and
rated companies deteriorate and earn lower ESG ratings how to weigh the importance of various pieces of data.
after such a short time, but rather that the managers of In this article, we will describe the path of refine-
the SR funds are swapping out companies apparently ment from the raw data points to the final scores, based on
with little regard for what the ESG ratings are telling the complete methodology detailed in the TRCRR Rule
them. We hope that as ESG ratings and similarly infor- Book [2014]. To make the description as clear as possible,
mative data gain prominence, they will become more we will follow three examples as they evolve from qualita-
wisely used. tive to quantitative, subjective to objective, raw to normal-
But in order to address those new inquiries on ized. You will gain an understanding of how the ratings
a large scale, we need objective measures. Negative address the challenges we have discussed. Although you
screening of objectionable products requires very little will see a number of equations provided for clarity’s sake,
research; either a company produces a product that a the procedures they represent will also be explained ver-
particular investor finds objectionable, or it does not. bally, so you will gain an intuitive grasp of the information,
Additionally, such basic product information is the even without having to parse mathematical expressions.
kind that companies invariably provide. But when we
start asking questions that could distinguish between ASSEMBLING THE RAW DATA
two companies making the same product, we soon run
into obscurity, opacity, and subjectivity where we need The most labor-intensive part of the process is
clarity, transparency, and objectivity. at the beginning, when Thomson Reuters ASSET4
Simply put, comparable ratings enable better com- (henceforth ASSET4) researchers collect more than 500
parisons. They allow researchers to study the statistical data points for each of some 5,000 companies. A typical
inf luence of ESG practices, and they allow investors company’s profile takes a week to assemble, drawing
to apply the researchers’ findings in ever-more-refined from a variety of sources including stock exchange fil-
strategies. They encourage companies to be more trans- ings, annual reports, and news outlets. These data points
parent in their practices, and they have the potential to are sorted at two levels, first into eighteen categories,
more than 250 key performance indicators (KPIs). First, To offer an analogy from linguistics, a KPI is to a mor-
note that this question requires a “yes” or “no” answer, pheme (the smallest meaningful unit of speech) as an
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
which makes it a Boolean rather than a metric question individual data point is to a phoneme (a mere sound).
(which might be answered either on a 1–5 scale or with Boolean KPIs are created using one of the fol-
an absolute number). Second, note that answering “yes” lowing conversion tables (Exhibit 1 being for positive
is not a good thing in this case. That means that this ques- polarity, and Exhibit 2 being for negative polarity):
tion is one of negative polarity (if a “yes” answer were a good Answering “yes” to a one-question KPI is scored
thing, then the question would have positive polarity). the same as answering “yes” to both questions in a
The second example is from the social pillar: two-question KPI. If the company answers “yes” to one
“Does the company report or show to use human and “no” to the other, the score is 0.5. Note how “not rel-
rights criteria in the selection or monitoring process evant” (NR) and “not available” (NA) are treated differ-
of its suppliers or sourcing partners? And does the com- ently. “Not relevant” means that the question is irrelevant
pany report or show to be ready to end a partnership
with a sourcing partner if human rights criteria are EXHIBIT 1
not met?” Positive Polarity KPI
Note that this example actually consists of two
distinct questions, but they will be lumped together into
a single KPI because the second question is dependent on
the first; a company that does not pay attention to human
rights when selecting its suppliers is not likely to drop
those suppliers for human rights-related reasons. But we
would not want to discard the second question, because
follow-through is an important factor to measure. (In the
next section, we will see how partial credit is given for
“yes/no” and “no/yes” answers to these dual-question
KPIs.) Also note that both of the questions are of positive
polarity, unlike that in the first example.
The third example is from the corporate governance EXHIBIT 2
pillar, and it requires simply the “percentage of shares held Negative Polarity KPI
by all insiders and 5% owners.” Hopefully, you recognize
this as a metric question, as opposed to a Boolean one.
This metric question in particular is asking for an absolute
number rather than a number on the relative 1–5 scale,
and converting so many disparate quantities into com-
parable ones will bring us to the first of many equations
in this article (but accompanied by accessible wording,
never fear!). And thankfully, due to the way the conver-
sion works, we do not need to be concerned whether this
question is of positive or negative polarity.
x I x I
information. We score this lack of an answer the same as
the lowest-scoring answer to that question given by any where x ∈ I tells us that when we are finding the
comparable company (that is, one in the same region, maximum reported value x, we are considering only
industry, or both). The logic behind this method is as the values reported by companies within the same
follows: If we scored NA as 0.5 like we do with NR, industry I. The symbol ∈ means “is an element of,”
then companies would have an easy way to game the so the expression as a whole is telling us that we are
ratings. For any question to which their answer would be taking the maximum value of x given that x is reported
negative, the company would be incentivized to refuse by a company in the same industry as the company that
to give any answer, thereby getting a 0.5 instead of a 0. reported xi, the individual value we are converting into
By scoring NA as the worst comparable score, we incen- a KPI score.
tivize companies to provide the information, but we also Why do we use the extra steps for environmental
do not punish them unduly for lacking it, as they can metric KPIs? Many KPIs in that pillar have low rates
never be the worst-scoring company as a result. of reporting, which would result in numerous scores of
We can easily see where our environmental and zero. To adjust for that and preserve comparability across
social pillar examples would fall in the previous exhibits. pillars, we set the NA score for metric KPIs at 0.4 instead
Although that is quite straightforward, the metric KPIs of 0 and use the additional steps for reported metrics, so
require some calculation to make the quantities compa- that reporting anything guarantees a company a score of
rable. Our corporate governance example will undergo higher than NA or NR, starting at 0.6 and maxing out
the following procedure: at 1. The NA scores being set at 0.4 gives the distribu-
tion a more normal shape (recall that the NR score is
1. Subtract the lowest reported number (among 0.5, halfway between the NA score and the minimum
regional peers) from the company’s reported score for a reported metric).
number. For the positive polarity version, we would simply
2. Subtract the lowest reported number from the remove the subtraction from one, and for an analogous
highest reported number (again, that’s the highest KPI in the corporate governance or social pillars, we
reported from among companies within the same would furthermore remove the multiplication by 0.4 and
geographical region). the addition to 0.6. To prevent any confusion, we have
3. Divide the result of (1) by the result of (2). provided a full list of the equations in the Appendix.
4. Subtract the result of (3) from one.
RELATIVE LEVELS OF IMPORTANCE
As you might expect, the above steps will vary
depending on the polarity of the question—if our metric The second step is to assign each KPI an RLI.
example had been of positive polarity, then we would RLIs range from 0 to 5 and are assigned based on sev-
have skipped (4). But there is also a twist for metric KPIs eral factors:
in the environmental pillar, which undergo two addi-
tional steps: the number—that is the quotient from (3), 1. How relevant that KPI is to the peer group (that
or if applicable, the difference from (4)—is multiplied is, companies within the same industry in the
5. How objectively measurable the KPI is. to adjust for the differences in reporting across industries
6. How robustly the statistical results for the KPI can and ensure proper peer-to-peer comparisons.
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
be validated using the academic literature. Most of the factors map onto one within-pillar
category each, except for two of the factors (7 and 10)
As you can see, assigning the RLIs involves some that are represented in all three categories of the envi-
degree of subjective judgment and a great deal of simpli- ronmental pillar (resource reduction, emission reduc-
fication, as all these considerations condense into a single tion, and product innovation). Exhibit 3 shows all the
value. That said, there are no subjective judgments being factors.
made on a company-by-company basis; all of the above To calculate an environmental KPI’s weight, we
considerations translate into quantitative indicators (if add together all the modified RLIs of the KPIs within
they do not begin as such) and are applied consistently, a factor to get a divisor for that factor, divide the KPI’s
so we need not think of the RLI assignment as any RLI by the factor’s divisor, and finally multiply by the
more hazardous than, for example, the two additional factor’s weight, as follows:
mathematical steps used to convert metric KPIs in the
environmental pillar. Any arbitrariness is spread evenly Environmental KPI
P Weiight
across companies, so none is especially advantaged or ⎛ RLI
L ⎞
disadvantaged. =⎜ ⎟ ∗ Factor Weight (2)
⎝ ∑ ( RLI
L Multiiplier ) ⎠
CALCULATING THE WEIGHTS
We then dynamically scale the scores on the KPI
Once assigned, each RLI is modified by a multi- for the peer group, such that the best score is equal to 1
plier that is based on the percentage of companies in the and the worst score is equal to 0.
peer group that gave an answer for the KPI in question. For the social and governance pillars, we likewise
The multiplier can take one of three values: 0, 0.5, or add the RLIs for the KPIs in each peer group to create
1. The percentage thresholds that determine the multi- divisors for the respective peer groups and divide the
plier vary by pillar; the environmental pillar has lower
reporting percentages in general compared with the
other two, so its thresholds are adjusted downward.
EXHIBIT 3
Ten Factors of the Environmental Pillar
• Weights in the Environmental Pillar: 0 (for when
less than 10% of the companies in the peer group
gave answers to the KPI), 0.5 (for when between
10% and 30% gave answers), or 1 (for when more
than 30% gave answers).
• Weights in the Corporate Governance and Social
Pillars: 0 (for when less than 0.5% of the companies
in the peer group gave answers to the KPI), 0.5
(for when between 0.5% and 15% gave answers),
or 1 (for when more than 15% gave answers).
with the social and corporate governance KPIs, thereby average cube of the Z-score of the adjusted Z-score. This
creating three raw pillar scores that we will convert into is a lot to unpack, but it essentially means that we are
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
finalized ratings in the next section. trying to tell whether (and if so, to what degree) the dis-
Every KPI is benchmarked according to industry, tribution is distorted away from a perfect bell curve in
region, or the entire universe of companies. In other either the positive or negative direction. We need to cube
words, a company’s score on a given KPI may alterna- the Z-scores in order to preserve the distinction between
tively be compared with the scores of other companies positive and negative skews, because merely multiplying a
in the same industry or region, or simply all the com- negative value by itself would result in a positive value and
panies in the universe. Corporate governance KPIs are thereby blind us to the direction (see Equation 4):
all compared within regions; environmental KPIs are all
Skew = ⎛
compared within industries; and social KPIs are variously 1 ⎞
compared within regions, industries, and the universe. ⎝ n ⎠
n
(4)
FROM RAW SCORES TO RATINGS ∗ ∑ ( adjusted Z - (adjusted Z - )) / σ )
3
i =1
C
r ng = 0.5
Interim ratin 3
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
RLI
L
=⎜ ⎟ ∗ Factor Weight
We have seen how the TRCRR are constructed, ⎝ ∑ ( RLI
L Multiiplier ) ⎠
The Journal of Investing 2016.25.2:103-112. Downloaded from www.iijournals.com by CORNELL UNIVERSITY on 06/09/16.
a few ways they have been used, and some evidence for
the effectiveness of these and similar data-driven strate- For social or corporate governance:
gies. Hopefully, this article leaves you with an intuitive report
e ing n < 0.5%
% → multipl
0 5% i ier = 0
grasp of the logic behind the process. We also hope
that you now have enhanced confidence in the quality 0.5% ≤ report
e ing n ≤ 15% → m
multipl
ltipli = 0.5
i ier
of these ratings, such that you feel confident in using report
e ing n > 15%
% → multipl
i ier = 1
them to study how ESG practices inf luence stock per- Social or Corporate Governance KPI
P Weiight
formance, stability, popular perceptions, and countless
other variables. ⎛ RLI
L ⎞
=⎜ ⎟
⎝ ∑ ( RLI
L Multiiplier ) ⎠
APPENDIX
Converting Raw Scores to Ratings
EQUATIONS
Step 1
Metric Score Conversion
Z -score = ( raw score – e ) / σ ( raw
raw scores r scores )
For environmental, positive polarity:
Step 2
⎛ xi min i x ⎞
Score = 0.6 0.4 ∗ ⎜ x I
⎟
⎜⎝ max x min i x ⎟⎠ Right tail:
x I x I
If Z-score > 3, adjusted Z-score = max(Z-score ≤ 3)
For environmental, negative polarity:
Left tail:
⎛ ⎛ xi min i x ⎞⎞ If Z-score <−3, adjusted Z-score = min(Z-score ≥−3)
Score = 0.6 0.4 ∗ ⎜ 1 − ⎜ x I
⎟⎟
⎜⎝ ⎜⎝ max x min i x ⎟⎠ ⎟⎠
x I x I Step 3
Skew = ⎛ ⎞
For social or corporate governance, positive polarity: 1
⎝ n ⎠
xi min
i x n
Score = x R
∗ ∑ ( adjusted
d Z-- (adjusted
ddjusted Z-- )) /σ )3
max x min
i x
x R x R i =1
1 1
Composite ESG Rating = ( Environmental
E i l ) + (Social ) Renneboog, L., J. Ter Horst, and C. Zhang. “The Price
3 3 of Ethics and Stakeholder Governance: The Performance
1 of Socially Responsible Mutual Funds.” Journal of Corporate
+ (Corporate Governancr e) Finance, Vol. 14, No. 3 ( June 2008a), pp. 302–322.
3
Adler, T., and M. Kritzman. “The Cost of Socially Respon- To order reprints of this article, please contact Dewey Palmieri
sible Investing.” The Journal of Portfolio Management, Vol. 35, at dpalmieri@ iijournals.com or 212-224-3675.
No. 1 (2008), pp. 52-56.